#061

I released concurrency-1.8.1.0, adding a newTVarConc method to MonadConc:

-- | Create a @TVar@. This may be implemented differently for speed.
--
-- > newTVarConc = atomically . newTVar
--
-- @since 1.8.1.0
newTVarConc :: a -> m (TVar (STM m) a)
newTVarConc = atomically . newTVar

The IO instances uses newTVarIO.

This week had a bunch of little things:

• I made spelling suggestions only bold the corrected words, which required a small change in search-api to get Elasticsearch tell us what those words are.

• I fixed a nearly two-year-old issue with a two-line diff.

• I switched off one A/B test (shingles: results were sadly inconclusive) and switched on another (a change to how we rank spelling suggestions).

• I shrunk our search-api and Elasticsearch resources down to what they were before we recently scaled up.

• I made a red graph less red.

• And I removed an old dependency, making the code shorter while I was at it.

I was off on Thursday to go look at some flats in Rickmansworth. More on that later.

I read The Path of Daggers (by Robert Jordan), the eighth book in the Wheel of Time series.

I continued with my project to get GOV.UK running on Kubernetes, and have made some good progress this week:

• Concourse builds docker images for all the apps (PR#1)
• The frontend apps are running in production-mode against the live GOV.UK apis (PR#5)
• I’m using Amazon’s Elastic Kubernetes Service to manage the cluster (PR#8)

I’ve got some really neat stuff set up, for example:

• An app which requests storage will automatically provision and attach an EBS volume
• An app which accepts external traffic will automatically provision an ALB and set up a Route53 entry

I’ve started to work on the publishing apps, but I’ve not done much more than deploy content-store and a proof-of-concept MongoDB instance. I’d like to also deploy search-api, router-api, and router (with their respective databases); putting in some fake data; and seeing if it all works. Once that proof-of-concept has been tested, I’d then like to switch to using the AWS managed Elasticsearch and Document DB, rather than running them myself.

In the process of switching govuk-k8s to using EKS, I also switched it from nginx to Caddy, because it turns out nginx has some really strange ideas about hostnames:

This also switches from nginx to caddy because nginx has some interesting ideas about hostnames:

1. If given statically, nginx resolves all hostnames at start-up.

   • If a hostname doesn’t resolve, nginx doesn’t start. This is a problem because the ALBs won’t exist until the apps are deployed.

   • If a hostname is changed to point to a different IP, nginx won’t pick up that change either.

2. If given dynamically (in a variable), nginx forces you to construct the proxy URL yourself.

   • It seems exceedingly difficult to construct a URL based on user-supplied input which doesn’t open a possibility of SSRF.

Caddy solves both of these problems by just behaving sensibly with hostnames.

I’ve known about Caddy for a while, but not tried it before. The configuration is way simpler than nginx, as it has more sensible defaults. I’m planning to switch all my servers from nginx to Caddy now.

I went to see three flats in Rickmansworth on Thursday, and decided to rent one of them. It’s a five minute walk from the tube station, which itself is about an hour away from the station next to work. My criteria when looking for a flat were:

• No more than an hour and a half from work
• The commute to work is a single tube journey / bus ride / whatever
• Can get at least 300Mb internet (preferably Hyperoptic, but that proved far too limiting)
• Has space for: VR gaming, table-top gaming, and a computer desk

My current flat is pretty small, and I’ve had to move my computer every time I wanted to use the table for boardgames, which sucks a lot. The flat I picked is over twice the size, and has a second bedroom which I can turn into a home office.